Reference: 'Parallel Neuronal Mechanisms for Short-Term Memory,' Earl K.
Miller and Robert Desimone, Science, 263:520-522, 28 January, 1994.
This is an interesting article describing the mechanisms of STM that the
authors have been investigating at NIMH over the last few years. For
attendees of the 1992 Appalachian Conference, this paper complements
Robert Desimone's presentation (which Paul Prueitt and I reported on
in Neuroscience last year). The key point is that the authors have
demonstrated two parallel STM mechanisms in monkey vision. The first
involves a category of neurons in the IT that are organized topologically
by location on the retina. These apparently receive input from feature-
matchers that are also organized topologically. The connecting synapses
apparently habituate easily and reversibly. Hence, if there is nothing
novel in the visual scene, these IT neurons do not fire. The habituation
lasts for a while, which allows for movement of the scene on the retina
and periodic motions of scene elements. The IT neurons output to a
topologically-organized array that merges the novelty signal with the
visual signal to designate scene patches to be attended to. In addition,
there is a capacity for monitoring the output of this array to determine
if it has not detected novelty in a selected region. That is used in
detecting repetition of scene elements.
The second mechanism uses STP in IT neurons to search for and recognize
known (non-novel) scene elements. This appears to involve a mechanism
similar to that used in the olfactory system--reafferent data used to
create a synthetic image for training the array, which then conducts a
search for matches in the visual signal. This is apparently controlled by
the prefrontal lobe. The IT neurons are arrayed topologically so that a
match can be passed to another topologically-organized array to merge the
match signal with the visual signal to produce a patch to be attended to.
This mechanism is not the preferred solution if detection of non-novelty
is enough to do the job. (Monkeys first use non-novelty detection and then
move to known-object detection.)
Note that both of these processes operate early in the visual processing
cascade (about 80-90 msec). They apparently operate continuously, but
patching their signals into consciousness is usually deliberate. This
begins to suggest to me that most scene processing in the brain is
continuous, with consciousness selecting this or that output stream to be
attended to. In other words, 'qualia' are there continuously, but we're
usually oblivious to them.
Internet: herwin at gmu.edu or erwin at trwacs.fp.trw.com
Working on Katchalsky networks....